Then again, multiple fronts are a good tactic while attacking anyway, specially when your opponent can't see all the units you have on each front.
Artificial Intelligence
- Thread starter Valgua
- Start date
Then again, multiple fronts are a good tactic while attacking anyway, specially when your opponent can't see all the units you have on each front.
On the contrary, I think huge stacks are MUCH easier for the AI to manage than 1upt units. But I hope for the best.
Caeserportugal
Prince
I think one of the screenshots of Azazell he was being atttacked by an AI though I can't remember. I think he was playing Egypt and he was attacked by Persia or Siam. I don't remember quite well.
Psyringe
Scout
I wouldn't expect the AI to be top notch out of the box. No matter how long they've developed and tested the AI, nothing can fully prepare it for a million of players worldwide hammering at it. What we can expect from Firaxis is that they'll keep improving the AI with expansions and patches. I also hope that the AI remains moddable, as the AI modding in Civ4 has brought huge improvements to the game, many of which were then incorporated into the official patches.
I also agree with AriochIV that 1upt units are harder to manage for the AI than stacks. The reason is simply that one of the most important tactical decisions is in which sequence the units should attack (sacrifice a weaker unit to ensure an easier victory for the stronger one, risking a slightly lower success chance for one combat when it means a higher gain for the next unit in the stack, etc.). This is a difficult decision that the Civ4 AI never really got right, since it wasn't capable to predict the success chances of battles between whole stacks. However, removing the stacks doesn't remove the problem of having to choose a good attack sequence, so the main reason for the Civ4 AI's weakness with stacks isn't addressed by this change at all. The 1upt rule additionally introduces the necessity of having to plan unit positioning better. Planning the movement of many 1upt units requires more forethought than calculating attack chances of units from a single stack, and forethought is the hardest part of AI programming, so I certainly expect it to be harder. (Actually I'm afraid that there will be little planning at all in that regard apart from a simple list of movement priorities, but I could very well be wrong here.)
The layered AI system sounds quite promising though, it's addressing one of the main weaknesses of the Civ4 AI: the lack of an actual plan to win the game. AI mods have alleviated this somewhat, but the Civ4 AI still makes far too many inefficient ad-hoc decisions, especially when moving units. Any improvement in that regard will certainly be welcome.
I also agree with AriochIV that 1upt units are harder to manage for the AI than stacks. The reason is simply that one of the most important tactical decisions is in which sequence the units should attack (sacrifice a weaker unit to ensure an easier victory for the stronger one, risking a slightly lower success chance for one combat when it means a higher gain for the next unit in the stack, etc.). This is a difficult decision that the Civ4 AI never really got right, since it wasn't capable to predict the success chances of battles between whole stacks. However, removing the stacks doesn't remove the problem of having to choose a good attack sequence, so the main reason for the Civ4 AI's weakness with stacks isn't addressed by this change at all. The 1upt rule additionally introduces the necessity of having to plan unit positioning better. Planning the movement of many 1upt units requires more forethought than calculating attack chances of units from a single stack, and forethought is the hardest part of AI programming, so I certainly expect it to be harder. (Actually I'm afraid that there will be little planning at all in that regard apart from a simple list of movement priorities, but I could very well be wrong here.)
The layered AI system sounds quite promising though, it's addressing one of the main weaknesses of the Civ4 AI: the lack of an actual plan to win the game. AI mods have alleviated this somewhat, but the Civ4 AI still makes far too many inefficient ad-hoc decisions, especially when moving units. Any improvement in that regard will certainly be welcome.
I'll point out that Panzer general had pretty decent AI especially on the defense. It was typically competent on the offense at the higher levels.
Most importantly the game with hands done the absolute best AI is chess. Which has one unit per stacking and relies on brute forcing to achieve a big chunk of its advantage over humans. Chess AI actually understand sacrificing so there is hope.
Most importantly the game with hands done the absolute best AI is chess. Which has one unit per stacking and relies on brute forcing to achieve a big chunk of its advantage over humans. Chess AI actually understand sacrificing so there is hope.
With the addition of ranged combat, plus the fact that you likely only have a few units in melee range, I don't see where sacrificing weaker units is going to be as prevalent in 5 versus 4 (or rather you will have fewer combinations available during each combat round).
One thing mentioned in by Firaxis but not delved into during the previews is how the "swapping adjacent units" mechanic is going to work. I've also seen it mentioned (though I forget where) that the 1UPT rule applies only at the end-of-turn so that mid-turn you can have units stacked. I can imagine it is possible to "stalemate" a unit in this case but if true would make movement for the AI and player much easier to manage - especially with mounted units that can dart though the front-line to attack once in range.
Comparing a 64 square Chess AI with finite pieces and no "economic" layer to a 1000+ square Civ AI is not fair or useful...
One thing mentioned in by Firaxis but not delved into during the previews is how the "swapping adjacent units" mechanic is going to work. I've also seen it mentioned (though I forget where) that the 1UPT rule applies only at the end-of-turn so that mid-turn you can have units stacked. I can imagine it is possible to "stalemate" a unit in this case but if true would make movement for the AI and player much easier to manage - especially with mounted units that can dart though the front-line to attack once in range.
Comparing a 64 square Chess AI with finite pieces and no "economic" layer to a 1000+ square Civ AI is not fair or useful...
Palance
Warlord
- Joined
- Oct 28, 2005
- Messages
- 100
That's true but don't forget that chess has had competitive AI for a long time (25 years? More?) With today's hardware they *should* be able to develop AI exponetially better and hopefully that power will be able to handle the exponentially more complex Civ model.
Getting somewhat OT but...
It is, and that is why Civ AI puts up even a fighting chance against most people and does pretty well against even more if they do not intentionally target the pattern behaviors.
I do think you underestimate the degree of exponential increase adding multiple opponent AIs, terrain, near-infinite unit possibilities, map-size, diplomacy, economics, production, time-limitations, and multiple "victory" conditions has on comparing the complexity of Civ to Chess.
Palance
Warlord
- Joined
- Oct 28, 2005
- Messages
- 100
Because of the three layered AI approach, a computer driven Civ won't have it's front lines pondering fiscal policy in between turns.
Units will only be affected by terrain, number of opponents, enemy units and so on. And only these conditions that are within it's scope, what it can see.
Units will only be affected by terrain, number of opponents, enemy units and so on. And only these conditions that are within it's scope, what it can see.
Gre_Magus
Manet
- Joined
- Oct 26, 2005
- Messages
- 273
The full game of Civ is exponentially more complex than chess - there's no comparison. At the highest level, there's little hope for an AI to use "brute force" to predict future turns. At that level, the AI should be like a human player - try to maximize production on a turn by turn level but then have "goals" or "scripts" that determine whether the focus should be on expanding, building up an army and conquering an opponent, increasing science, culture, etc.
In terms of combat, I agree that 1 unit-per-hex make combat a more difficult problem for the computer to solve than stacks, but I hope this just encourages the programmers to make the AI more competent.
In Civ 4 (and Gal Civ 2, for that matter) the AI seemed to over emphasize defense because (as far as I could tell) it had little or no ability to analyze maneuver. From the human players' perspective, it almost always made sense to create one or at most two stacks and attack one city at a time, unless he had an overwhelming advantage over the opponent. With enough warning, the AI would try to attack that stack, reinforce that one city, or, in the best of cases, launch a counter attack. But in almost all cases, it kept at least 4 or more defenders in all of its other cities, limiting the size of its defensive stack. The player might only leave 1 or 2 defenders in each city, and so, all else being equal, could always bring a bigger stack to any given battle.
By allowing cities to defend themselves and forcing units out of them, the AI in Civ 5 should be more challenging, even if it is not that much "smarter" than the AI in Civ 4. It should bring more of its units to a given battle. Where the Civ 5 AI could go very wrong is if it leaves too many units back to "defend" and allows the player to defeat its army piecemeal. I'd rather beat an AI that over commits and allows me to flank it by maneuvering through an allied civ or something than one that never brings its whole army to bear because there's a hard-coded minimal defender rule.
The same thing goes for rushes. I'd rather see an AI rush me with warriors at 3000 BC then be coded to only expand during the "expansion phase". If they manage to beat me, great. If not, at least it was cool!
kozzer
Liberated Autocrat
- Joined
- Dec 8, 2005
- Messages
- 141
As a programmer, the way I understand it is that there would be no unit passing. The terms 'strategic', 'tactical', etc are merely labels. It's really based on conceptual levels.
There are 4 levels/implementations of AI in the game:
Level 1: "Grand Strategic AI" - picks a victory condition, and informs Level 2
Level 2: "Strategic AI" - Manages the empire as a whole, city management, diplomacy,etc
Level 3: "Operational AI" - Manages the units in total aka the entire war front (probably setting up chokepoint/city defense)
Level 4: "Tactical AI" - Controls all individual units
So, control of units will always be under the domain of level 4, never any other level. The level above informs the level below, and otherwise they're totally separate. And in general terms with regard to programming, having multiple tools with specific purposes (Civ5 AI model) is much more effective and efficient than 1 tool that does everything (Civ4 AI model). So I'm optimistic.
Of course, this is all speculation on my part, but I thought I'd throw my (not completely uninformed) 2 cents in.
My understanding of Thyrwyn's post is that The Operational AI may sub-divide the units into "Western Front" and "Eastern Front" and each front would have its own "Tactical AI" that directs the units within its Operational front; almost like you see currently within HoI:3 (the WWII game). Such a sub-division strategy is probably but whether it is being used we have no idea (but it is not precluded by the list you quote).
If it does then the operational AI would need to be able to reallocate existing units between sub-Fronts in order to operate effectively - but would be a complex trade-off mechanic (sacrifice the east for the west or put the east into a "holding action" while the west attacks).
AutomatedTeller
Frequent poster
The problem with building a "competant" AI is that it's not going to be able to learn, which radically limits it's ability to deal with the human.
if no human ever shared information about how to beat the AI, we would have much, much higher opinions of how the AI does in Civ. But with places like the war academy, exploits show up much faster.
btw - humans do stupid things, too. In my first ever PBEM game, I had 2 coastal cities burned on the first turn of the invasion by seaborne invasion of maces, because they were protected by warriors. What's worst is that I was a protective civ and one of them had walls and a castle - I had longbows.. heck, i had muskets.
Later in that same game, I burned an opponents coastal city in much the same way.
if no human ever shared information about how to beat the AI, we would have much, much higher opinions of how the AI does in Civ. But with places like the war academy, exploits show up much faster.
btw - humans do stupid things, too. In my first ever PBEM game, I had 2 coastal cities burned on the first turn of the invasion by seaborne invasion of maces, because they were protected by warriors. What's worst is that I was a protective civ and one of them had walls and a castle - I had longbows.. heck, i had muskets.
Later in that same game, I burned an opponents coastal city in much the same way.
kozzer
Liberated Autocrat
- Joined
- Dec 8, 2005
- Messages
- 141
Yeah, if that's what Thyrwyn was getting at, then I misunderstood his post. I dunno about the split between "fronts" whether that's a physical split (i.e. different instances of the operational AI), a logical split within a single operational AI, or if there is no split at all. I guess we'll have to see how it plays out.
If there are any sort of splits, I really cannot imagine how the AI would determine when a split should occur or "where" the split should be. Unless, of course, it's based on the opponent. But if I open a 2-front war (attacking with 2 groups of units, one from the west, one front the east) will the AI recognize that as 2 distinct "fronts"? Or if 2 civs attack a 3rd on the same side, will the AI see that as 2 fronts or 1?
It's those questions that cause me to think that there's only one operational AI that acts like the Field Marshall taking orders from the Strategic AI (who would be the government in this analogy). The OpAI would order the tactical AI to "defend that city" or "defend this side of the river" which would then actually go about moving the proper units into the proper tiles (optimizing for terrain, etc).
So, yeah, I'm not really contradicting anyone, just sorta thinking onto the keyboard.
playshogi
Emperor
- Joined
- Nov 1, 2001
- Messages
- 1,479
The thing I saw was the AI leaving a catapult exposed and easily killed. The narrator even said it was a bad idea to do that. Ranged combat seems devastating, how can you defend against it?
OT but...
Fight fire with fire; you need your own ranged combat units.
Attacking a defesive position filled with ranged weapons IS difficult; and should make conquering territory much more difficult than in Civ 4.
One thing that would be nice is if you could use your infantry to negate the Zone-of-Control of the opponent so that you could rush mounted units through gaps in the defensive lines. Basically your troops engage the enemy so they cannot detach units to stop the mounted charge. Make it an actual action as opposed to normal behavior so that you cannot both "hold" and "attack" at the same time.
If an AI could do this effectively that would be great since it might actually be added...
It also seems to me that they're making the AI realize that they are playing a game, that they should try and win.
In every previous installment, the AI acts ideologically, or stupidly. What I mean is that they'll hate you just because you've chosen monarchy, and not for any tangible reason, really. Or they'll hate you for doing something that was simply wildly out of your comprehension that they would hate you for it. For instance, I once met up with france while coming up to the renessance era; just getting boats that can travel across the seas, and they immediately went to war with me because I had so effectively traded around with people he hated, and wasn't his religion. That's kind of stupid, both from a gameplay perspective, and also from any perspective that benefits him. He spent so much time and resources attacking me that the civs around him just basically absorbed him.
What I'm expecting to see is an AI that understands the tools it has, and understands the rules of the game. It will see that you have more iron then it, and see that it has a stronger offensive position currently, and calculate that it is a good idea to go to war with you, rather then just off-handedly deciding it should be so.
Also, the AI will certainly have goals, more then any other installment. The British will be naval, it would be stupid if they weren't. The Greeks will pamper to the City-States, it would be stupid not to. And we, the players, will know this about them, and be able to act in kind, just as the AI will act against us in a specific way depending on what Civilization we are.
If we're playing as the Aztecs, it would be STUPID of the AI to think we AREN'T going to war. A smart AI will simply ASSUME that the Aztecs, who get bonuses for killing non-barbarian units, WILL CERTAINLY go to war with them. A stupid AI will not, and will die quickly.
That's my largest hope for the franchise, is that all players, AI or otherwise, will always understand what the other players strengths and weaknesses are, and defend or attack as needed. It would be incredibly rediculous if the AI spawned as the Egyptians, and never attempted to build Wonders. It would ALSO be incredibly rediculous if the PLAYER spawned as the Egyptians and never attempted to build wonders, and the AI should know that. The AI should, and hopefully will, always be programmed to assume the other player is playing to his Civilization's strength.
If that can happen then we'll FINALLY be playing civilization as a GAME and not as a simulator.
Psyringe
Scout
I'd like to point out that, if this happens, the franchise will lose a great deal of its appeal to me. I love Civ precisely because it allows me to indulge into the feeling of an unfolding history of a parallel world, instead of just playing a game. For that reason, I want some rivals to act like Louis in your description. For that reason, I also would not want the AI to act upon pre-game knowledge like "the Aztecs will probably attack me" - I want such notions to emerge from the developing history of the world, and not from some pre-programmed out-of-game knowledge (I'd also be afraid that this would shoehorn games into being more identical to each other).
One of the hallmarks of the Civ franchise has always been that it combined elements of strategy games and simulations, and allowed for both playing styles. Many people (me included) roleplay their games to some extent, i.e. we deliberately take strategically suboptimal decisions because they wouldn't be in character for the behavior of our civilization. For example, we might let a rival live (although killing him off might be better) because he helped us in the past. Or we go to total war on a remote civilization because it betrayed our ally. Civ allows for that playing style because to the some extent the AI plays that way too so players are not at a disadvantage (compared to the AI) when roleplaying a bit in their decisions. If the franchise ever decides to drop the simulation aspects in favor for a pure "gaming" approach, then these elements would be lost.
Of course, AI improvements don't necessarily mean a shift towards a "gaming" approach. Having a grand plan will benefit the AI in either approach, and the AI can (and hopefully will) keep the balance between "game" and "simulation" just as well.
I'd just like you to realize that the approach which you deride in capitals as "STUPID", is entirely enjoyable, believable, in-character, and welcome to many other players. Thanks.
This is an entirely valid point, but I believe that a strong, goal oriented AI will provide more opportunity for such roleplay, and not less.
1) The AI will have a personality before the game even starts. This means that you have a stronger relationship with the AI simply because he is a "character" and pretty strong one. His gameplay attributes and rules will make him a stronger presence in the game, from a role play perspective.
2) It's easier to role play if the other player's goals are clear and make sense. If a player attacks you simply "Because" then it's harder to remain connected to the sence of a real, breathing world. If one is tied strongly to the idea of irrationality ruling the day, there are City states that are literally labled as "Irrational", which may facilitate that desire.
In the end, I imagine there will be amble opportunity to have the kind of role play experience you are talking about and have it make sense. If I were to meet the Aztecs, for the first time, and they immediately went to war with me, I would not think that is stupid. I would think "Yup, that's Montezuma for you!" It makes sense for his character, AND it's to his benefit to do so.
If I were to meet Ghandi and he immediately went to war with me because I had traded with his enemies before I even knew of his existence, then I'm left scratching my head. The illusion breaks when these kinds of things happen, and I strongly believe that a system that ensures that everyone is pursuing goals will only serve to strengthen any kind of roleplay that will occur.
Jonolith, it's not entirely true. In particular the AI changes that were made with BtS made the AI more "victory aware". I think we've all had a "what the..." moment when an AI has pulled off a cultural victory in some obscure corner of the globe. But in general, yes the civ4/warlords/bts AIs were there mostly to roleplay and provide what could be described as an "obstacle" to your victory.
Personally I prefer the angle of making the AIs play to win the game (though with some limitations, like not dogpiling the human player on turn 10 at Deity!). At least for me, I think that improves the long-term enjoyment of the game.
The worst thing that could happen is to do a civrev, and beat the Deity difficulty on the second play. The absence of a challenge is a bigger problem than the absence of immersion, IMO. So far it sounds like civ5 is shifting the balance a bit further in the direction I prefer.
Personally I prefer the angle of making the AIs play to win the game (though with some limitations, like not dogpiling the human player on turn 10 at Deity!). At least for me, I think that improves the long-term enjoyment of the game.
The worst thing that could happen is to do a civrev, and beat the Deity difficulty on the second play. The absence of a challenge is a bigger problem than the absence of immersion, IMO. So far it sounds like civ5 is shifting the balance a bit further in the direction I prefer.
